This is a very small treeherder alternative developped as a pet project to monitor the divvun CI pipelines. It has a very very small subset of what treeherder actually does and is tailored to my need although I'm open to suggestions.
To deploy this, you can run the docker container at
ghcr.io/eijebong/berger:main
, you'll need to provide it the following
environment variables:
TASKCLUSTER_ROOT_URL
: The root URL of your taskcluster instance, example:https://taskcluster.your.org
AMQP_ADDR
: The address of the rabbitmq used for taskcluster's pulse, example:amqp://user:[email protected]
DATABASE_URL
: The URL of the postgresql database berger will use to store its data, example:postgres://user@[email protected]/berger
REDIRECT_URL
(optional): If your instance is not deployed asberger.${TASKCLUSTER_ROOT_URL}
, set this toberger.url/auth/callback
GITHUB_TRIGGER_HOOK
(optional): The hook to call to trigger jobs for github repositories. Example:project/github-trigger
. More on that in theTrigger hook
section.
To use the login feature, you have to add an oauth client definition to your taskcluster instance.
Example:
registered_clients: [ {"clientId": "berger", "responseType": "code", "scope": ["berger:*", "hooks:trigger-hook:*], "whitelisted": true, "redirectUri": ["https://berger.taskcluster.url/auth/callback"], "maxExpires": "1 year" } ]
To avoid leaking repositories, berger uses scopes to filter what's displayed.
You should add a berger:get-repo:org/name
scope to roles that should have
access to those repositories. You can also replace org/name
by a *
to allow
all repositories.
This will pick up every task that's created on your taskcluster instance, there's currently no filtering. It'll assume two things about created tasks.
First, the source of the decision task should be github's ${event.compare}
.
Second, you should add a tag with the current ref named git_ref
.
metadata:
source: ${event.compare}
...
tags:
git_ref: ${event.ref}
...
This is due to taskcluster not having any way to link a task to a push event
except for the source and custom metadata like tags.
Technically the git_ref
is not necessary to link to a push but then you
cannot differentiate a branch push and a tag push as the source would show up
as the same. Note that I'm open to any way of simplificating these requirements.
Berger includes a way to run jobs from a predefined hook. This is useful to
avoid requiring empty commits to rerun some CI. It works by triggering the hook
specified by GITHUB_TRIGGER_HOOK
with the following arguments:
- repo_org
- repo_name
- branch
You can then craft a hook that runs your decision task for a push for that repo. The hooks should have the following trigger schema:
type: object
properties:
branch:
type: string
repo_org:
type: string
repo_name:
type: string
additionalProperties: false